Solor-TXT-7B-Ultra is a fine-tuned version of Qwen2.5-7B-Instruct, optimized for deep reasoning and multi-turn conversation. This model is trained to think before it speaks, utilizing a dedicated <thought> block to process complex context and strategy.
Model Features
Advanced Reasoning: Injected with specialized thought-patterns to improve logic in long conversations.
Multi-Turn Mastery: Trained on a filtered subset of UltraChat 200k for stable, high-quality dialogue.
Architect: Fine-tuned by DatOneStormyz using Unsloth on an NVIDIA A100.
Suggested System Prompt
To trigger the reasoning behavior effectively, use the following system prompt:
"You are Solor-TXT. You must think deeply inside tags before every response, especially in long conversations."
